Scalable prediction of global online media news virality

Abstract

News reports shape the public perception of the critical social, political, and economical events around the world. Yet, the way in which emergent phenomena are reported in the news makes the early prediction of such phenomena a challenging task. We propose a scalable community-based probabilistic framework to model the spreading of news about events in online media. Our approach exploits the latent community structure in the global news media and uses the affiliation of the early adopters with a variety of communities to identify the events widely reported in the news at the early stage of their spread. The time complexity of our approach is linear in the number of news reports. It is also amenable to efficient parallelization. To demonstrate these features, the inference algorithm is parallelized for message passing paradigm and tested on the Rensselaer Polytechnic Institute Advanced Multiprocessing Optimized System, one of the fastest Blue Gene/Q supercomputers in the world. Thanks to the community-level features of the early adopters, the model gains an improvement of 20% in the early detection of the most massively reported events compared with the feature-based machine learning algorithm. Its parallelization scheme achieves orders of magnitude speedup.